Allow Multipart hdf5 Datasets

[genematx]

This enables support for consolidating datasets that consist of multiple hdf5 files. The HDF5Adapter in Tiled already can accept multiple uris and concatenate them into a single array. However, if the filenames are passed in the Resource document in the form of a template, HDF5Consolidator did not have the ability to register these files per each Datum (as done for sequences of tiff or jpeg file in MultipartRelatedConsolidator).

This PR declares a new MultipartHDF5Consolidator to be used for datasets of "application/x-hdf5" mimetype, but only if the Resource document declares a "template" in resource_kwargs (or parameters in StreamResource). Otherwise, the usual HDF5Consolidator is used for the same mimetype.

Related discussion in Mattermost: https://mattermost.hzdr.de/bluesky/pl/uhpt8h54bpn9iyoxpb65tupr9y

[codecov]

Codecov Report

❌ Patch coverage is 94.28571% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 68.90%. Comparing base (77301fa) to head (117fe58).

Files with missing lines	Patch %	Lines
src/bluesky_tiled_plugins/writing/consolidators.py	83.33%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #19      +/-   ##
==========================================
+ Coverage   68.83%   68.90%   +0.07%     
==========================================
  Files          14       14              
  Lines        1832     1833       +1     
==========================================
+ Hits         1261     1263       +2     
+ Misses        571      570       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[prjemian]

Here's one of the cases I want to test, with different values for num_acquire & num_images (num_exposures does not result in a different number of acquired frames).

acquire_time = 0.01
gap_between_frames = 0.001
num_acquire = 2  # num arg to count plan
num_exposures = 1  # number of camera exposures per frame (not supported by all detectors)
num_images = 3  # number of frames per acquire

adsimdet.cam.stage_sigs["acquire_time"] = acquire_time
adsimdet.cam.stage_sigs["image_mode"] = "Multiple"
adsimdet.cam.stage_sigs["num_exposures"] = num_exposures
adsimdet.cam.stage_sigs["num_images"] = num_images

adsimdet.hdf1.stage_sigs["num_capture"] = num_images
adsimdet.hdf1.stage_sigs.move_to_end("capture", last=True)

I'm expecting a 4-D dataset: (2, 3, nx, ny) but the HDF5 file data has shape (3, nx, ny). Inspecting how bluesky.plans.count works, the detector is staged, then iterates over num_acquire. The HDF5 plugin is not staged for additional acquisitions.

Not a situation to be fixed here, but one of the challenges in reviewing.

[prjemian]

Acquisition results in a single resource and two datum docs.

(uid,) = RE(  # noqa
    bp.count(  # noqa
        [adsimdet],
        num_acquire,
        md=dict(
            title="Area Detector with default HDF5 File Name",
            purpose="image"),
        )
    )

run = cat[uid]

but tiled client fails to get the dataset:

dataset = run.primary.read()

(10 tries resulting in HTTPStatusError: Server error '500 Internal Server Error').

[prjemian]

Setting num_acquire = 1 and selecting any number of images results in data that tiled.client can provide from the HDF5 file, with this branch.